hcl/json/scanner/scanner_test.go
Paul Hinze feb701804f Restore old behavior for \" within interpolations
Before the parser rewrite, HCL would silently convert `\"` within
interpolation braces to `"`.

After the conversion, this became a syntax error.

We've found several instances of Terraform configs in the wild using
this syntax. It results in a hard "syntax error" message during config
parsing. While avoiding the extra escape on double quotes within
interpolations is definitely preferred, the UX of the syntax error feels
harsh enough to be worth inserting this backwards compatibility for now,
leaving us the option of deprecating it with a warning down the line.
2015-11-19 12:04:02 -06:00

364 lines
8.6 KiB
Go

package scanner
import (
"bytes"
"fmt"
"testing"
"github.com/hashicorp/hcl/json/token"
)
var f100 = "ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff"
type tokenPair struct {
tok token.Type
text string
}
var tokenLists = map[string][]tokenPair{
"operator": []tokenPair{
{token.LBRACK, "["},
{token.LBRACE, "{"},
{token.COMMA, ","},
{token.PERIOD, "."},
{token.RBRACK, "]"},
{token.RBRACE, "}"},
},
"bool": []tokenPair{
{token.BOOL, "true"},
{token.BOOL, "false"},
},
"string": []tokenPair{
{token.STRING, `" "`},
{token.STRING, `"a"`},
{token.STRING, `"本"`},
{token.STRING, `"${file("foo")}"`},
{token.STRING, `"${file(\"foo\")}"`},
{token.STRING, `"\a"`},
{token.STRING, `"\b"`},
{token.STRING, `"\f"`},
{token.STRING, `"\n"`},
{token.STRING, `"\r"`},
{token.STRING, `"\t"`},
{token.STRING, `"\v"`},
{token.STRING, `"\""`},
{token.STRING, `"\000"`},
{token.STRING, `"\777"`},
{token.STRING, `"\x00"`},
{token.STRING, `"\xff"`},
{token.STRING, `"\u0000"`},
{token.STRING, `"\ufA16"`},
{token.STRING, `"\U00000000"`},
{token.STRING, `"\U0000ffAB"`},
{token.STRING, `"` + f100 + `"`},
},
"number": []tokenPair{
{token.NUMBER, "0"},
{token.NUMBER, "1"},
{token.NUMBER, "9"},
{token.NUMBER, "42"},
{token.NUMBER, "1234567890"},
{token.NUMBER, "-0"},
{token.NUMBER, "-1"},
{token.NUMBER, "-9"},
{token.NUMBER, "-42"},
{token.NUMBER, "-1234567890"},
},
"float": []tokenPair{
{token.FLOAT, "0."},
{token.FLOAT, "1."},
{token.FLOAT, "42."},
{token.FLOAT, "01234567890."},
{token.FLOAT, ".0"},
{token.FLOAT, ".1"},
{token.FLOAT, ".42"},
{token.FLOAT, ".0123456789"},
{token.FLOAT, "0.0"},
{token.FLOAT, "1.0"},
{token.FLOAT, "42.0"},
{token.FLOAT, "01234567890.0"},
{token.FLOAT, "0e0"},
{token.FLOAT, "1e0"},
{token.FLOAT, "42e0"},
{token.FLOAT, "01234567890e0"},
{token.FLOAT, "0E0"},
{token.FLOAT, "1E0"},
{token.FLOAT, "42E0"},
{token.FLOAT, "01234567890E0"},
{token.FLOAT, "0e+10"},
{token.FLOAT, "1e-10"},
{token.FLOAT, "42e+10"},
{token.FLOAT, "01234567890e-10"},
{token.FLOAT, "0E+10"},
{token.FLOAT, "1E-10"},
{token.FLOAT, "42E+10"},
{token.FLOAT, "01234567890E-10"},
{token.FLOAT, "01.8e0"},
{token.FLOAT, "1.4e0"},
{token.FLOAT, "42.2e0"},
{token.FLOAT, "01234567890.12e0"},
{token.FLOAT, "0.E0"},
{token.FLOAT, "1.12E0"},
{token.FLOAT, "42.123E0"},
{token.FLOAT, "01234567890.213E0"},
{token.FLOAT, "0.2e+10"},
{token.FLOAT, "1.2e-10"},
{token.FLOAT, "42.54e+10"},
{token.FLOAT, "01234567890.98e-10"},
{token.FLOAT, "0.1E+10"},
{token.FLOAT, "1.1E-10"},
{token.FLOAT, "42.1E+10"},
{token.FLOAT, "01234567890.1E-10"},
{token.FLOAT, "-0.0"},
{token.FLOAT, "-1.0"},
{token.FLOAT, "-42.0"},
{token.FLOAT, "-01234567890.0"},
{token.FLOAT, "-0e0"},
{token.FLOAT, "-1e0"},
{token.FLOAT, "-42e0"},
{token.FLOAT, "-01234567890e0"},
{token.FLOAT, "-0E0"},
{token.FLOAT, "-1E0"},
{token.FLOAT, "-42E0"},
{token.FLOAT, "-01234567890E0"},
{token.FLOAT, "-0e+10"},
{token.FLOAT, "-1e-10"},
{token.FLOAT, "-42e+10"},
{token.FLOAT, "-01234567890e-10"},
{token.FLOAT, "-0E+10"},
{token.FLOAT, "-1E-10"},
{token.FLOAT, "-42E+10"},
{token.FLOAT, "-01234567890E-10"},
{token.FLOAT, "-01.8e0"},
{token.FLOAT, "-1.4e0"},
{token.FLOAT, "-42.2e0"},
{token.FLOAT, "-01234567890.12e0"},
{token.FLOAT, "-0.E0"},
{token.FLOAT, "-1.12E0"},
{token.FLOAT, "-42.123E0"},
{token.FLOAT, "-01234567890.213E0"},
{token.FLOAT, "-0.2e+10"},
{token.FLOAT, "-1.2e-10"},
{token.FLOAT, "-42.54e+10"},
{token.FLOAT, "-01234567890.98e-10"},
{token.FLOAT, "-0.1E+10"},
{token.FLOAT, "-1.1E-10"},
{token.FLOAT, "-42.1E+10"},
{token.FLOAT, "-01234567890.1E-10"},
},
}
var orderedTokenLists = []string{
"comment",
"operator",
"bool",
"string",
"number",
"float",
}
func TestPosition(t *testing.T) {
// create artifical source code
buf := new(bytes.Buffer)
for _, listName := range orderedTokenLists {
for _, ident := range tokenLists[listName] {
fmt.Fprintf(buf, "\t\t\t\t%s\n", ident.text)
}
}
s := New(buf.Bytes())
pos := token.Pos{"", 4, 1, 5}
s.Scan()
for _, listName := range orderedTokenLists {
for _, k := range tokenLists[listName] {
curPos := s.tokPos
// fmt.Printf("[%q] s = %+v:%+v\n", k.text, curPos.Offset, curPos.Column)
if curPos.Offset != pos.Offset {
t.Fatalf("offset = %d, want %d for %q", curPos.Offset, pos.Offset, k.text)
}
if curPos.Line != pos.Line {
t.Fatalf("line = %d, want %d for %q", curPos.Line, pos.Line, k.text)
}
if curPos.Column != pos.Column {
t.Fatalf("column = %d, want %d for %q", curPos.Column, pos.Column, k.text)
}
pos.Offset += 4 + len(k.text) + 1 // 4 tabs + token bytes + newline
pos.Line += countNewlines(k.text) + 1 // each token is on a new line
s.Error = func(pos token.Pos, msg string) {
t.Errorf("error %q for %q", msg, k.text)
}
s.Scan()
}
}
// make sure there were no token-internal errors reported by scanner
if s.ErrorCount != 0 {
t.Errorf("%d errors", s.ErrorCount)
}
}
func TestComment(t *testing.T) {
testTokenList(t, tokenLists["comment"])
}
func TestOperator(t *testing.T) {
testTokenList(t, tokenLists["operator"])
}
func TestBool(t *testing.T) {
testTokenList(t, tokenLists["bool"])
}
func TestIdent(t *testing.T) {
testTokenList(t, tokenLists["ident"])
}
func TestString(t *testing.T) {
testTokenList(t, tokenLists["string"])
}
func TestNumber(t *testing.T) {
testTokenList(t, tokenLists["number"])
}
func TestFloat(t *testing.T) {
testTokenList(t, tokenLists["float"])
}
func TestRealExample(t *testing.T) {
complexReal := `
{
"variable": {
"foo": {
"default": "bar",
"description": "bar",
"depends_on": ["something"]
}
}
}`
literals := []struct {
tokenType token.Type
literal string
}{
{token.LBRACE, `{`},
{token.STRING, `"variable"`},
{token.COLON, `:`},
{token.LBRACE, `{`},
{token.STRING, `"foo"`},
{token.COLON, `:`},
{token.LBRACE, `{`},
{token.STRING, `"default"`},
{token.COLON, `:`},
{token.STRING, `"bar"`},
{token.COMMA, `,`},
{token.STRING, `"description"`},
{token.COLON, `:`},
{token.STRING, `"bar"`},
{token.COMMA, `,`},
{token.STRING, `"depends_on"`},
{token.COLON, `:`},
{token.LBRACK, `[`},
{token.STRING, `"something"`},
{token.RBRACK, `]`},
{token.RBRACE, `}`},
{token.RBRACE, `}`},
{token.RBRACE, `}`},
{token.EOF, ``},
}
s := New([]byte(complexReal))
for _, l := range literals {
tok := s.Scan()
if l.tokenType != tok.Type {
t.Errorf("got: %s want %s for %s\n", tok, l.tokenType, tok.String())
}
if l.literal != tok.Text {
t.Errorf("got: %s want %s\n", tok, l.literal)
}
}
}
func TestError(t *testing.T) {
testError(t, "\x80", "1:1", "illegal UTF-8 encoding", token.ILLEGAL)
testError(t, "\xff", "1:1", "illegal UTF-8 encoding", token.ILLEGAL)
testError(t, `"ab`+"\x80", "1:4", "illegal UTF-8 encoding", token.STRING)
testError(t, `"abc`+"\xff", "1:5", "illegal UTF-8 encoding", token.STRING)
testError(t, `01238`, "1:7", "numbers cannot start with 0", token.NUMBER)
testError(t, `01238123`, "1:10", "numbers cannot start with 0", token.NUMBER)
testError(t, `'aa'`, "1:1", "illegal char: '", token.ILLEGAL)
testError(t, `"`, "1:2", "literal not terminated", token.STRING)
testError(t, `"abc`, "1:5", "literal not terminated", token.STRING)
testError(t, `"abc`+"\n", "1:5", "literal not terminated", token.STRING)
}
func testError(t *testing.T, src, pos, msg string, tok token.Type) {
s := New([]byte(src))
errorCalled := false
s.Error = func(p token.Pos, m string) {
if !errorCalled {
if pos != p.String() {
t.Errorf("pos = %q, want %q for %q", p, pos, src)
}
if m != msg {
t.Errorf("msg = %q, want %q for %q", m, msg, src)
}
errorCalled = true
}
}
tk := s.Scan()
if tk.Type != tok {
t.Errorf("tok = %s, want %s for %q", tk, tok, src)
}
if !errorCalled {
t.Errorf("error handler not called for %q", src)
}
if s.ErrorCount == 0 {
t.Errorf("count = %d, want > 0 for %q", s.ErrorCount, src)
}
}
func testTokenList(t *testing.T, tokenList []tokenPair) {
// create artifical source code
buf := new(bytes.Buffer)
for _, ident := range tokenList {
fmt.Fprintf(buf, "%s\n", ident.text)
}
s := New(buf.Bytes())
for _, ident := range tokenList {
tok := s.Scan()
if tok.Type != ident.tok {
t.Errorf("tok = %q want %q for %q\n", tok, ident.tok, ident.text)
}
if tok.Text != ident.text {
t.Errorf("text = %q want %q", tok.String(), ident.text)
}
}
}
func countNewlines(s string) int {
n := 0
for _, ch := range s {
if ch == '\n' {
n++
}
}
return n
}